Hitachi Energy develops AI factory power simulation

Hitachi Energy has developed a power-system simulation capability using NVIDIA Omniverse libraries, allowing engineers to evaluate proposed AI data centre electrical infrastructure before physical equipment is installed. The initial application concentrates on emerging 800V DC distribution architectures, where increasing computing density creates additional requirements for power conversion, system resilience and coordination with electricity networks.

The simulation environment combines electrical engineering models with scenarios representing the changing power requirements of AI computing equipment. It is intended to support comparisons between alternative infrastructure designs, including the arrangement of equipment from the utility grid connection through to the computing racks.

The new capability forms part of the NVIDIA DSX Sim partner ecosystem and extends Hitachi Energy’s existing collaboration with NVIDIA on AI factory infrastructure. Hitachi had previously demonstrated an 800V DC power and control architecture integrated with NVIDIA’s reference design during March 2026, while the October development introduces a further simulation capability for evaluating proposed systems.

AI computing facilities can impose electrical demands that differ from those of conventional data centres. Large concentrations of accelerator hardware create high power density, while changes in computing workload can cause variations in consumption that must be accommodated by the supporting electrical infrastructure.

The scale and timing of those variations depend on the computing architecture, workload management and installed equipment. Electrical infrastructure must therefore be designed for both sustained operation and the conditions that arise as load changes, rather than relying solely on an estimate of average electricity demand.

At the utility interface, transformers and switchgear must accommodate the intended connection capacity and applicable fault conditions. Downstream equipment distributes electricity towards the computing systems, with power conversion, backup supplies and control arrangements providing the electrical characteristics required by the connected loads.

These components operate as an interconnected system. A change in the assumed computing load can affect equipment ratings, voltage performance, thermal losses and the design of protective arrangements elsewhere in the electrical chain.

Hitachi Energy’s simulation environment is intended to represent those interactions before procurement and construction decisions are finalised. Engineering teams can apply different load profiles and operating conditions to assess how a proposed configuration responds, then compare the results with alternative arrangements.

The initial focus on 800V DC distribution reflects the increasing interest in using higher DC voltages to supply dense computing installations. For a given power transfer, increasing voltage reduces the current required, provided the other electrical conditions remain comparable.

Lower current can reduce resistive losses in conductors because those losses increase with the square of current. It may also permit different conductor arrangements, although insulation requirements, protection equipment, conversion efficiency and other design constraints must be considered alongside conductor losses.

The overall efficiency of an electrical distribution system depends on the complete power chain. Reducing one category of loss does not establish that a new architecture is more efficient under every operating condition, particularly where the design introduces different power conversion stages or equipment operating points.

Higher-voltage DC distribution also changes protection requirements. Unlike alternating current, direct current does not pass through a natural current zero during each electrical cycle, making interruption of fault current a different engineering problem for switching equipment.

Protection arrangements must therefore be selected for the relevant DC voltage, prospective fault current and fault clearing requirements. Equipment isolation, maintenance procedures and coordination between converters and protective devices all influence whether the proposed architecture can operate safely.

Simulation provides a means of examining these constraints before the physical infrastructure is constructed. Engineers can investigate how load changes, equipment outages and selected fault scenarios affect the proposed system, while identifying assumptions that require further testing.

Resilience assessment is another application. A system may be designed with multiple power supply paths, backup equipment or alternative operating modes, but its actual resilience depends on how those elements behave when individual components become unavailable.

Models can help compare those arrangements under defined scenarios. However, simulated performance remains dependent on the quality of the equipment models and the accuracy of the assumptions used to represent real operating conditions.

The shared environment uses OpenUSD workflows and NVIDIA Omniverse libraries to bring engineering information together. Hitachi Energy contributes its electrical system expertise and models, while the simulation infrastructure provides a common environment for evaluating alternative designs and their behaviour.

An AI assistant is also integrated into the interface, allowing users to explore the digital model through natural language, retrieve information and compare simulation results. This provides an additional method for interacting with engineering data, although the validity of the underlying calculations remains separate from the convenience of the interface.

Engineering judgement and established verification procedures remain necessary when interpreting simulation results. A model can indicate how equipment is expected to respond within its assumptions, but physical testing, equipment certification and commissioning are still required to establish performance in an installed system.

Hitachi Energy has associated the development with efforts to improve the number of AI computing tokens produced for each watt consumed. That provides a potential optimisation measure linking electrical infrastructure with computing performance, but no independently verified increase in tokens per watt has been disclosed for a specific installation using the October capability.

The wider electricity demand associated with data centres gives the work a substantial engineering context. The International Energy Agency reports that global data centre electricity consumption increased by approximately 17% during 2025, with consumption at AI-focused facilities growing more quickly.

Increasing electrical demand places pressure on grid connection capacity as well as the design of infrastructure within individual facilities. Earlier evaluation may identify equipment and system constraints, although modelling alone cannot increase the physical capacity available from an electricity network or remove the need for connection approvals.

Hitachi Energy has also received NVIDIA DSX Ready BESS qualification for its battery energy storage offering. Storage can form part of a power infrastructure design, providing additional operational flexibility within its power and energy limits, but no specific battery configuration is required or validated by the new simulation announcement.

The company’s development is therefore best understood as an additional engineering assessment capability rather than a commercially demonstrated improvement in infrastructure efficiency. The next measurable evidence will concern its application to specific projects, the accuracy of model predictions and the operating results obtained when simulated power systems are built and tested.